Potential of seasonal hydrological forecasting of monthly run-offvolumes for the Rhone and Arve Rivers from April to July

Abstract. In a context where water management is becoming increasingly important, reliable seasonal forecasting of discharge in rivers is crucial for making decisions several months in advance. This paper explores the potential of seasonal forecasting of run-off volumes produced by ensemble streamflow forecasting using past climatology and comparing it to the more commonly used average of past discharge measurements. The seasonal forecast was obtained for the Arve and Rhone rivers by simulation using the Routing System model for lead times of 30, 90 and 120 days. The initialization was performed on a validated simulation of 12 and 16 years for the Arve and Rhone rivers, respectively, obtained through long-term calibration. The performance was assessed by indicators called accuracy and thinness . The normalized mean average error (NMAE) was used to compare the performance of the seasonal forecast with the average of the past measurements. After a bias correction of the seasonal forecast of the Rhone River with the observed run-off volumes during the different lead times, the correlation of the median forecast with the measurements (accuracy) was larger than 0.55 for all lead times from April to July. The Arve River's accuracy was improved by disregarding the year 2007 member, leading to the floods of the 3rd and 9th of July, for lead times of 90 and 120 days. This resulting in the period of April to July having correlation accuracies higher than 0.5. For both rivers, the 80 % confidence interval of the seasonal forecast was relatively thin compared to the measurements (thinness) for the months of April to July. The NMAE was used to validate the range of validity of the forecast. The correction of the forecast resulted in more months being favorable for seasonal forecasting for the Rhone River. The post-processing on the Arve River decreased the difference between the measurement and the forecast (NMAE). Further investigation should concentrate on dividing the meteorological datasets to produce a strong median forecast and confidence interval